1 /*
2 * linux/include/linux/sunrpc/svc.h
3 *
4 * RPC server declarations.
5 *
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 */
8
9
10 #ifndef SUNRPC_SVC_H
11 #define SUNRPC_SVC_H
12
13 #include <linux/in.h>
14 #include <linux/in6.h>
15 #include <linux/sunrpc/types.h>
16 #include <linux/sunrpc/xdr.h>
17 #include <linux/sunrpc/auth.h>
18 #include <linux/sunrpc/svcauth.h>
19 #include <linux/wait.h>
20 #include <linux/mm.h>
21
22 /* statistics for svc_pool structures */
23 struct svc_pool_stats {
24 atomic_long_t packets;
25 unsigned long sockets_queued;
26 atomic_long_t threads_woken;
27 atomic_long_t threads_timedout;
28 };
29
30 /*
31 *
32 * RPC service thread pool.
33 *
34 * Pool of threads and temporary sockets. Generally there is only
35 * a single one of these per RPC service, but on NUMA machines those
36 * services that can benefit from it (i.e. nfs but not lockd) will
37 * have one pool per NUMA node. This optimisation reduces cross-
38 * node traffic on multi-node NUMA NFS servers.
39 */
40 struct svc_pool {
41 unsigned int sp_id; /* pool id; also node id on NUMA */
42 spinlock_t sp_lock; /* protects all fields */
43 struct list_head sp_sockets; /* pending sockets */
44 unsigned int sp_nrthreads; /* # of threads in pool */
45 struct list_head sp_all_threads; /* all server threads */
46 struct svc_pool_stats sp_stats; /* statistics on pool operation */
47 #define SP_TASK_PENDING (0) /* still work to do even if no
48 * xprt is queued. */
49 unsigned long sp_flags;
50 } ____cacheline_aligned_in_smp;
51
52 struct svc_serv;
53
54 struct svc_serv_ops {
55 /* Callback to use when last thread exits. */
56 void (*svo_shutdown)(struct svc_serv *, struct net *);
57
58 /* function for service threads to run */
59 int (*svo_function)(void *);
60
61 /* queue up a transport for servicing */
62 void (*svo_enqueue_xprt)(struct svc_xprt *);
63
64 /* set up thread (or whatever) execution context */
65 int (*svo_setup)(struct svc_serv *, struct svc_pool *, int);
66
67 /* optional module to count when adding threads (pooled svcs only) */
68 struct module *svo_module;
69 };
70
71 /*
72 * RPC service.
73 *
74 * An RPC service is a ``daemon,'' possibly multithreaded, which
75 * receives and processes incoming RPC messages.
76 * It has one or more transport sockets associated with it, and maintains
77 * a list of idle threads waiting for input.
78 *
79 * We currently do not support more than one RPC program per daemon.
80 */
81 struct svc_serv {
82 struct svc_program * sv_program; /* RPC program */
83 struct svc_stat * sv_stats; /* RPC statistics */
84 spinlock_t sv_lock;
85 unsigned int sv_nrthreads; /* # of server threads */
86 unsigned int sv_maxconn; /* max connections allowed or
87 * '0' causing max to be based
88 * on number of threads. */
89
90 unsigned int sv_max_payload; /* datagram payload size */
91 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */
92 unsigned int sv_xdrsize; /* XDR buffer size */
93 struct list_head sv_permsocks; /* all permanent sockets */
94 struct list_head sv_tempsocks; /* all temporary sockets */
95 int sv_tmpcnt; /* count of temporary sockets */
96 struct timer_list sv_temptimer; /* timer for aging temporary sockets */
97
98 char * sv_name; /* service name */
99
100 unsigned int sv_nrpools; /* number of thread pools */
101 struct svc_pool * sv_pools; /* array of thread pools */
102 struct svc_serv_ops *sv_ops; /* server operations */
103 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
104 struct list_head sv_cb_list; /* queue for callback requests
105 * that arrive over the same
106 * connection */
107 spinlock_t sv_cb_lock; /* protects the svc_cb_list */
108 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no
109 * entries in the svc_cb_list */
110 struct svc_xprt *sv_bc_xprt; /* callback on fore channel */
111 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
112 };
113
114 /*
115 * We use sv_nrthreads as a reference count. svc_destroy() drops
116 * this refcount, so we need to bump it up around operations that
117 * change the number of threads. Horrible, but there it is.
118 * Should be called with the "service mutex" held.
119 */
svc_get(struct svc_serv * serv)120 static inline void svc_get(struct svc_serv *serv)
121 {
122 serv->sv_nrthreads++;
123 }
124
125 /*
126 * Maximum payload size supported by a kernel RPC server.
127 * This is use to determine the max number of pages nfsd is
128 * willing to return in a single READ operation.
129 *
130 * These happen to all be powers of 2, which is not strictly
131 * necessary but helps enforce the real limitation, which is
132 * that they should be multiples of PAGE_SIZE.
133 *
134 * For UDP transports, a block plus NFS,RPC, and UDP headers
135 * has to fit into the IP datagram limit of 64K. The largest
136 * feasible number for all known page sizes is probably 48K,
137 * but we choose 32K here. This is the same as the historical
138 * Linux limit; someone who cares more about NFS/UDP performance
139 * can test a larger number.
140 *
141 * For TCP transports we have more freedom. A size of 1MB is
142 * chosen to match the client limit. Other OSes are known to
143 * have larger limits, but those numbers are probably beyond
144 * the point of diminishing returns.
145 */
146 #define RPCSVC_MAXPAYLOAD (1*1024*1024u)
147 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD
148 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u)
149
150 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
151
152 /*
153 * RPC Requsts and replies are stored in one or more pages.
154 * We maintain an array of pages for each server thread.
155 * Requests are copied into these pages as they arrive. Remaining
156 * pages are available to write the reply into.
157 *
158 * Pages are sent using ->sendpage so each server thread needs to
159 * allocate more to replace those used in sending. To help keep track
160 * of these pages we have a receive list where all pages initialy live,
161 * and a send list where pages are moved to when there are to be part
162 * of a reply.
163 *
164 * We use xdr_buf for holding responses as it fits well with NFS
165 * read responses (that have a header, and some data pages, and possibly
166 * a tail) and means we can share some client side routines.
167 *
168 * The xdr_buf.head kvec always points to the first page in the rq_*pages
169 * list. The xdr_buf.pages pointer points to the second page on that
170 * list. xdr_buf.tail points to the end of the first page.
171 * This assumes that the non-page part of an rpc reply will fit
172 * in a page - NFSd ensures this. lockd also has no trouble.
173 *
174 * Each request/reply pair can have at most one "payload", plus two pages,
175 * one for the request, and one for the reply.
176 * We using ->sendfile to return read data, we might need one extra page
177 * if the request is not page-aligned. So add another '1'.
178 */
179 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
180 + 2 + 1)
181
svc_getnl(struct kvec * iov)182 static inline u32 svc_getnl(struct kvec *iov)
183 {
184 __be32 val, *vp;
185 vp = iov->iov_base;
186 val = *vp++;
187 iov->iov_base = (void*)vp;
188 iov->iov_len -= sizeof(__be32);
189 return ntohl(val);
190 }
191
svc_putnl(struct kvec * iov,u32 val)192 static inline void svc_putnl(struct kvec *iov, u32 val)
193 {
194 __be32 *vp = iov->iov_base + iov->iov_len;
195 *vp = htonl(val);
196 iov->iov_len += sizeof(__be32);
197 }
198
svc_getu32(struct kvec * iov)199 static inline __be32 svc_getu32(struct kvec *iov)
200 {
201 __be32 val, *vp;
202 vp = iov->iov_base;
203 val = *vp++;
204 iov->iov_base = (void*)vp;
205 iov->iov_len -= sizeof(__be32);
206 return val;
207 }
208
svc_ungetu32(struct kvec * iov)209 static inline void svc_ungetu32(struct kvec *iov)
210 {
211 __be32 *vp = (__be32 *)iov->iov_base;
212 iov->iov_base = (void *)(vp - 1);
213 iov->iov_len += sizeof(*vp);
214 }
215
svc_putu32(struct kvec * iov,__be32 val)216 static inline void svc_putu32(struct kvec *iov, __be32 val)
217 {
218 __be32 *vp = iov->iov_base + iov->iov_len;
219 *vp = val;
220 iov->iov_len += sizeof(__be32);
221 }
222
223 /*
224 * The context of a single thread, including the request currently being
225 * processed.
226 */
227 struct svc_rqst {
228 struct list_head rq_all; /* all threads list */
229 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */
230 struct svc_xprt * rq_xprt; /* transport ptr */
231
232 struct sockaddr_storage rq_addr; /* peer address */
233 size_t rq_addrlen;
234 struct sockaddr_storage rq_daddr; /* dest addr of request
235 * - reply from here */
236 size_t rq_daddrlen;
237
238 struct svc_serv * rq_server; /* RPC service definition */
239 struct svc_pool * rq_pool; /* thread pool */
240 struct svc_procedure * rq_procinfo; /* procedure info */
241 struct auth_ops * rq_authop; /* authentication flavour */
242 struct svc_cred rq_cred; /* auth info */
243 void * rq_xprt_ctxt; /* transport specific context ptr */
244 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */
245
246 size_t rq_xprt_hlen; /* xprt header len */
247 struct xdr_buf rq_arg;
248 struct xdr_buf rq_res;
249 struct page * rq_pages[RPCSVC_MAXPAGES];
250 struct page * *rq_respages; /* points into rq_pages */
251 struct page * *rq_next_page; /* next reply page to use */
252 struct page * *rq_page_end; /* one past the last page */
253
254 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
255
256 __be32 rq_xid; /* transmission id */
257 u32 rq_prog; /* program number */
258 u32 rq_vers; /* program version */
259 u32 rq_proc; /* procedure number */
260 u32 rq_prot; /* IP protocol */
261 int rq_cachetype; /* catering to nfsd */
262 #define RQ_SECURE (0) /* secure port */
263 #define RQ_LOCAL (1) /* local request */
264 #define RQ_USEDEFERRAL (2) /* use deferral */
265 #define RQ_DROPME (3) /* drop current reply */
266 #define RQ_SPLICE_OK (4) /* turned off in gss privacy
267 * to prevent encrypting page
268 * cache pages */
269 #define RQ_VICTIM (5) /* about to be shut down */
270 #define RQ_BUSY (6) /* request is busy */
271 #define RQ_DATA (7) /* request has data */
272 unsigned long rq_flags; /* flags field */
273
274 void * rq_argp; /* decoded arguments */
275 void * rq_resp; /* xdr'd results */
276 void * rq_auth_data; /* flavor-specific data */
277 int rq_auth_slack; /* extra space xdr code
278 * should leave in head
279 * for krb5i, krb5p.
280 */
281 int rq_reserved; /* space on socket outq
282 * reserved for this request
283 */
284
285 struct cache_req rq_chandle; /* handle passed to caches for
286 * request delaying
287 */
288 /* Catering to nfsd */
289 struct auth_domain * rq_client; /* RPC peer info */
290 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */
291 struct svc_cacherep * rq_cacherep; /* cache info */
292 struct task_struct *rq_task; /* service thread */
293 spinlock_t rq_lock; /* per-request lock */
294 };
295
296 #define SVC_NET(svc_rqst) (svc_rqst->rq_xprt->xpt_net)
297
298 /*
299 * Rigorous type checking on sockaddr type conversions
300 */
svc_addr_in(const struct svc_rqst * rqst)301 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
302 {
303 return (struct sockaddr_in *) &rqst->rq_addr;
304 }
305
svc_addr_in6(const struct svc_rqst * rqst)306 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
307 {
308 return (struct sockaddr_in6 *) &rqst->rq_addr;
309 }
310
svc_addr(const struct svc_rqst * rqst)311 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
312 {
313 return (struct sockaddr *) &rqst->rq_addr;
314 }
315
svc_daddr_in(const struct svc_rqst * rqst)316 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
317 {
318 return (struct sockaddr_in *) &rqst->rq_daddr;
319 }
320
svc_daddr_in6(const struct svc_rqst * rqst)321 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
322 {
323 return (struct sockaddr_in6 *) &rqst->rq_daddr;
324 }
325
svc_daddr(const struct svc_rqst * rqst)326 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
327 {
328 return (struct sockaddr *) &rqst->rq_daddr;
329 }
330
331 /*
332 * Check buffer bounds after decoding arguments
333 */
334 static inline int
xdr_argsize_check(struct svc_rqst * rqstp,__be32 * p)335 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
336 {
337 char *cp = (char *)p;
338 struct kvec *vec = &rqstp->rq_arg.head[0];
339 return cp >= (char*)vec->iov_base
340 && cp <= (char*)vec->iov_base + vec->iov_len;
341 }
342
343 static inline int
xdr_ressize_check(struct svc_rqst * rqstp,__be32 * p)344 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
345 {
346 struct kvec *vec = &rqstp->rq_res.head[0];
347 char *cp = (char*)p;
348
349 vec->iov_len = cp - (char*)vec->iov_base;
350
351 return vec->iov_len <= PAGE_SIZE;
352 }
353
svc_free_res_pages(struct svc_rqst * rqstp)354 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
355 {
356 while (rqstp->rq_next_page != rqstp->rq_respages) {
357 struct page **pp = --rqstp->rq_next_page;
358 if (*pp) {
359 put_page(*pp);
360 *pp = NULL;
361 }
362 }
363 }
364
365 struct svc_deferred_req {
366 u32 prot; /* protocol (UDP or TCP) */
367 struct svc_xprt *xprt;
368 struct sockaddr_storage addr; /* where reply must go */
369 size_t addrlen;
370 struct sockaddr_storage daddr; /* where reply must come from */
371 size_t daddrlen;
372 struct cache_deferred_req handle;
373 size_t xprt_hlen;
374 int argslen;
375 __be32 args[0];
376 };
377
378 /*
379 * List of RPC programs on the same transport endpoint
380 */
381 struct svc_program {
382 struct svc_program * pg_next; /* other programs (same xprt) */
383 u32 pg_prog; /* program number */
384 unsigned int pg_lovers; /* lowest version */
385 unsigned int pg_hivers; /* highest version */
386 unsigned int pg_nvers; /* number of versions */
387 struct svc_version ** pg_vers; /* version array */
388 char * pg_name; /* service name */
389 char * pg_class; /* class name: services sharing authentication */
390 struct svc_stat * pg_stats; /* rpc statistics */
391 int (*pg_authenticate)(struct svc_rqst *);
392 };
393
394 /*
395 * RPC program version
396 */
397 struct svc_version {
398 u32 vs_vers; /* version number */
399 u32 vs_nproc; /* number of procedures */
400 struct svc_procedure * vs_proc; /* per-procedure info */
401 u32 vs_xdrsize; /* xdrsize needed for this version */
402
403 unsigned int vs_hidden : 1, /* Don't register with portmapper.
404 * Only used for nfsacl so far. */
405 vs_rpcb_optnl:1;/* Don't care the result of register.
406 * Only used for nfsv4. */
407
408 /* Override dispatch function (e.g. when caching replies).
409 * A return value of 0 means drop the request.
410 * vs_dispatch == NULL means use default dispatcher.
411 */
412 int (*vs_dispatch)(struct svc_rqst *, __be32 *);
413 };
414
415 /*
416 * RPC procedure info
417 */
418 typedef __be32 (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
419 struct svc_procedure {
420 svc_procfunc pc_func; /* process the request */
421 kxdrproc_t pc_decode; /* XDR decode args */
422 kxdrproc_t pc_encode; /* XDR encode result */
423 kxdrproc_t pc_release; /* XDR free result */
424 unsigned int pc_argsize; /* argument struct size */
425 unsigned int pc_ressize; /* result struct size */
426 unsigned int pc_count; /* call count */
427 unsigned int pc_cachetype; /* cache info (NFS) */
428 unsigned int pc_xdrressize; /* maximum size of XDR reply */
429 };
430
431 /*
432 * Mode for mapping cpus to pools.
433 */
434 enum {
435 SVC_POOL_AUTO = -1, /* choose one of the others */
436 SVC_POOL_GLOBAL, /* no mapping, just a single global pool
437 * (legacy & UP mode) */
438 SVC_POOL_PERCPU, /* one pool per cpu */
439 SVC_POOL_PERNODE /* one pool per numa node */
440 };
441
442 struct svc_pool_map {
443 int count; /* How many svc_servs use us */
444 int mode; /* Note: int not enum to avoid
445 * warnings about "enumeration value
446 * not handled in switch" */
447 unsigned int npools;
448 unsigned int *pool_to; /* maps pool id to cpu or node */
449 unsigned int *to_pool; /* maps cpu or node to pool id */
450 };
451
452 extern struct svc_pool_map svc_pool_map;
453
454 /*
455 * Function prototypes.
456 */
457 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
458 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
459 int svc_bind(struct svc_serv *serv, struct net *net);
460 struct svc_serv *svc_create(struct svc_program *, unsigned int,
461 struct svc_serv_ops *);
462 struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv,
463 struct svc_pool *pool, int node);
464 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
465 struct svc_pool *pool, int node);
466 void svc_rqst_free(struct svc_rqst *);
467 void svc_exit_thread(struct svc_rqst *);
468 unsigned int svc_pool_map_get(void);
469 void svc_pool_map_put(void);
470 struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int,
471 struct svc_serv_ops *);
472 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
473 int svc_set_num_threads_sync(struct svc_serv *, struct svc_pool *, int);
474 int svc_pool_stats_open(struct svc_serv *serv, struct file *file);
475 void svc_destroy(struct svc_serv *);
476 void svc_shutdown_net(struct svc_serv *, struct net *);
477 int svc_process(struct svc_rqst *);
478 int bc_svc_process(struct svc_serv *, struct rpc_rqst *,
479 struct svc_rqst *);
480 int svc_register(const struct svc_serv *, struct net *, const int,
481 const unsigned short, const unsigned short);
482
483 void svc_wake_up(struct svc_serv *);
484 void svc_reserve(struct svc_rqst *rqstp, int space);
485 struct svc_pool * svc_pool_for_cpu(struct svc_serv *serv, int cpu);
486 char * svc_print_addr(struct svc_rqst *, char *, size_t);
487
488 #define RPC_MAX_ADDRBUFLEN (63U)
489
490 /*
491 * When we want to reduce the size of the reserved space in the response
492 * buffer, we need to take into account the size of any checksum data that
493 * may be at the end of the packet. This is difficult to determine exactly
494 * for all cases without actually generating the checksum, so we just use a
495 * static value.
496 */
svc_reserve_auth(struct svc_rqst * rqstp,int space)497 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
498 {
499 svc_reserve(rqstp, space + rqstp->rq_auth_slack);
500 }
501
502 #endif /* SUNRPC_SVC_H */
503